repo-development
Maintain Prefect itself safely: repo layout, scoped AGENTS rules, focused validation, generated artifacts, and package-boundary caveats.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- repo-development
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Maintain Prefect itself safely: repo layout, scoped AGENTS rules, focused validation, generated artifacts, and package-boundary caveats.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/generated-artifacts.md6.6 KB
- 参考文档references/repo-layout.md8.1 KB
- 参考文档references/testing-and-validation.md6.9 KB
- 参考文档references/troubleshooting.md8.1 KB
- 脚本scripts/select_prefect_tests.py20.4 KB
03 SKILL.md 正文
Prefect Repo Development
Use this sub-skill when the task is to change the Prefect repository itself rather than author user flows, operate deployments, call the API, or run the CLI as an end user.
Route Here For
- Choosing source and test files for changes under
src/prefect/,tests/,docs/,client/,src/integrations/, orui-v2/. - Applying scoped
AGENTS.mdrules, including the source/test mirror, client/server boundaries, docs generation rules, integration boundaries, and UI-v2 workflow. - Selecting focused pytest, ruff, pre-commit, docs, generated-schema, or
prefect-clientvalidation commands before broader checks. - Maintaining generated files such as settings types, schemas, CLI docs, OpenAPI docs, API references, examples, and integration docs.
- Creating issue repro scripts under
repros/and avoiding maintainer anti-patterns such as bypassing server orchestration state transitions.
Route Elsewhere
- Use
flow-task-authoringfor user-facing@flow,@task, states, futures, task runners, caching, and result behavior. - Use
deployments-workersfor deployments,prefect.yaml, serve, workers, work pools, and run submission as a user workflow. - Use
cli-server-operationsfor operating the Prefect CLI, profiles, server startup/status, variables, artifacts, and diagnostics. - Use
api-client-settingsfor Python client usage, settings/profile APIs, schemas as a user, and theprefect-clientpackage distinction. - Use
events-blocks-assetsfor events, automations, blocks, assets, variables, and concurrency primitives as user-facing features.
Working Pattern
- Read the nearest scoped
AGENTS.mdbefore editing; deeper files override broader files. - Map changed source paths to mirrored tests and component-specific invariants before writing code.
- Keep behavior changes minimal, preserve public APIs unless explicitly approved, and update docs or generated artifacts when source-of-truth files change.
- Run the most focused safe validation first, then broaden only when the change affects shared behavior, schemas, docs, client packaging, or CI selection.
- Keep service-heavy, Docker, Postgres, integration, UI browser, and release/publish commands opt-in unless the task explicitly requires them.
References
- Start with repo-layout for source/test mapping, scoped instructions, package boundaries, and maintainer caveats.
- Use testing-and-validation to choose focused pytest, lint, pre-commit, CI matrix, and repro-script checks.
- Use generated-artifacts when schemas, settings, docs, API references, examples, UI service types, or
prefect-clientmetadata may be stale. - Use troubleshooting for missing tools, dependency drift, AGENTS conflicts, async test context failures, service requirements, and state-transition anti-patterns.
Bundled Helper
Run python scripts/select_prefect_tests.py --help from this sub-skill directory to inspect a safe test-selection helper. It accepts changed paths and prints suggested validation commands without running tests, starting services, reading git history, or mutating files.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).